Scribd
Upload
106 views

1.017/1.010 Class 15 Confidence Intervals: Interval Estimates

Parameter estimates computed from a random sample x1, x2,., xN can vary around the unknown true value a. For any given estimate, we seek a two-sided confidence interval that is likely to include the true value. An interval estimate is often derived from a standardized statistic.

Uploaded by

Dr. Ir. R. Didin Kusdian, MT.
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
106 views

1.017/1.010 Class 15 Confidence Intervals: Interval Estimates

Parameter estimates computed from a random sample x1, x2,., xN can vary around the unknown true value a. For any given estimate, we seek a two-sided confidence interval that is likely to include the true value. An interval estimate is often derived from a standardized statistic.

Uploaded by

Dr. Ir. R. Didin Kusdian, MT.
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

1.017/1.

010 Class 15
Confidence Intervals

Interval Estimates

Parameter estimates computed from a random sample x1, x2,..., xN can


vary around the unknown true value. For any given estimate, we seek a
two-sided confidence interval that is likely to include the true value a:

a L ≤ a ≤ aU

[aL, aU] is called an interval estimate.

Standardized Statistics

Interval estimate is often derived from a standardized statistic. This is a


random variable that depends on both the unknown true value a and its
estimate â .

An example is the z statistic:

aˆ − a
z (aˆ , a) =
SD(aˆ )

If the estimate is unbiased E[z] = 0 and Var[z] = 1, for any x or aˆ probability


distribution with finite moments (prove).

Example:

Suppose:

a = E[x] = mean of the x probability distribution


â = mx = sample mean (of random sample outcome x1, x2,..., xN).

Then:

m x − E[ x ] m x − E[ x ]
z (aˆ , a ) = =
SD(m x ) SD( x)
N

1
Deriving Interval Estimates

If we know the probability distribution of the standardized statistic z we


can derive an interval estimate. Specify the probability 1-α that z falls in
the interval [zL, zU] for a given value of a (e.g. 0.95):

P[ z L ≤ z (aˆ , a ) ≤ zU ] = 1 − α

Suppose that [zL, zU] is selected so that the probability that z lies above the
interval is the same as the probability that it lies below the interval. This
gives a two-sided interval [zL, zU] for z:

α
P[ z (aˆ , a) ≤ z L ] = Fz ( z L ) =
2

α
P[ z (aˆ , a) ≥ zU ] = 1 − Fz ( zU ) =
2

α   α
z L = FZ−1   zU = FZ−1 1 − 
2  2

From the Central Limit Theorem we know that â is normal and that z has a
unit normal distribution [i.e. z~ N(0,1)] for large sample sizes. In this
case the CDF Fz(z) and its inverse can be evaluated from standard normal
distribution tables or with the MATLAB functions normcdf and norminv.

If 1-α = 0.95 then zL = -1.96 and zU = +1.96.

Substitute the definition of z to obtain the corresponding two-sided interval


for aˆ :

aˆ − a
P[ z L ≤ ≤ zU ] = P[ a + z L SD[aˆ ] ≤ aˆ ≤ a + zU SD[aˆ ] ] = 1 − α
SD[aˆ ]

aˆ ≥ a + z L SD[aˆ ] aˆ ≤ a + zU SD[aˆ ]

Probability is 1-α that actual sample estimate â lies in this interval.

Now suppose that this relatively likely event occurs when the outcome of a
particular experiment (i.e. the â obtained from a particular random
sample) is aˆ . Then the true a must satisfy the following inequality:

aˆ − zU SD[aˆ ] ≤ a ≤ aˆ − z L SD[aˆ ]

2
This gives the desired 1-α confidence interval for a:

a L = aˆ − zU SD[aˆ ] aU = aˆ − z L SD[aˆ ]

We can obtain SD[ aˆ ] in two ways:

1. Derive directly from the definition of the estimator


aˆ = aˆ ( x1 , x 2 ,..., x N ) (not always possible). This usually requires
replacing population statistics [e.g. Std(x)] by sample statistics [e.g.
sx]

2. Stochastic simulation, using a = aˆ in random number


generator (usually possible but not exact). This generally requires
an assumption about the form of the underlying distribution Fx(x).

The confidence interval is wider for larger SD[ â ]

The confidence interval is wider for larger 1-α (e.g. 99%)

Summary

To derive a two-sided confidence interval:

1. Specify significance level α


2. Compute estimate â from the data
3. Compute Std [ â ] in one of two ways:
• If possible, relate Std( â ) to Std(x) and use the approximation
Std(x) ≈ sx (i.e assume the unknown population standard
deviation is equal to the sample standard deviation
computed from data).
• Otherwise, derive Std( aˆ ) using stochastic simulation
4. Compute zL and zU from and the specified α, assuming an
appropriate form for the CDF Fx(x)
5. Apply the two-sided confidence interval formula

Example – Large-sample two-sided confidence interval for the population mean:

Consider the sample mean mx, used to estimate the population mean E[x].
In this case, a = E[x] and aˆ = mx

Use result from Class 13 to derive SD[mx] directly, replacing SD[x] by the
sample standard deviation sx :

SD[ x ] sx
SD[m x ] = ≈
N N

3
So the large sample (assume z is normal) two-sided 95% confidence
interval for the population mean is:

sx sx
m x − 1.96 ≤ E[ x ] ≤ m x + 1.96
N N

Suppose:

[x1, x2,..., x10 ] = [ 0.1 2.9 1.0 1.4 0.23 0.54 1.57 8.0 0.40 1.6]

Then mx = 1.77, sx = 2.34, N = 10 and:

2.34 2.34
1.77 − 1.96 ≤ E[ x ] ≤ 1.77 + 1.96
10 10
0.32 ≤ E[ x ] ≤ 3.23

Suppose we don’t know the probability distribution of z (e.g. because the


sample size is too small to justify using the Central Limit Theorem) but we
know the distribution of x (except for a few unknown parameters).

Then we can approximate Fz(z) with a virtual experiment, replacing


unknown parameters with estimates computed from the random sample.
Once this is done we can plot Fz(z) vs z and identify from the plot the Fz-1
values needed to derive zL and zU. Usually we make a large sample
(normal) assumption.

Copyright 2003 Massachusetts Institute of Technology


Last modified Oct. 8, 2003

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy